The generalized power spectrum model (GPSM), which has been shown to account for a large number of psychoacoustic and speech intelligibility (SI) experiments, was extended to assess audio quality. Like the GPSM, the suggested audio quality model, GPSMq, combines features from the power spectrum model (PSM) and envelope power-spectrum model (EPSM). GPSMq utilizes signal-to-noise ratios (SNRs) in the power and envelope power domains to model the addition or removal of energy by the signal processing under test. Four audio quality databases that introduce linear and nonlinear distortions to music and speech signals were assessed to cover a large variety of distortions cases. GPSMq provided better overall prediction performance than other state-of-the-art auditory-model-based objective quality measures. The results demonstrate that the power and envelope power SNR metric is appropriate for predicting audio quality for a variety of signal distortions in addition to psychoacoustics and SI. This supports the notion that the auditory system extracts a universal set of auditory features to be analyzed in a task-dependent decision stage.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.